FAQ

Company
Question logo
What is your definition of credit fraud?
We build you a customized model based on your definition of credit fraud. How do you define credit fraud?
Question logo
How is your company exceptionally qualified to successfully perform your task?
With the Bleckwen credit fraud solution, we operate in a niche market without many agile competitors and are specialized enough to defend and maintain our position in the market. It is essential to mention that we do not replace traditional tools or internal data science teams. We complement existing processes to uncover this residual fraud that they can not. We have the best team equipped to succeed. In Bleckwen, we have a diverse international group of world-class data scientists with expertise in artificial intelligence and machine learning. AI is our core competency, and we have built a solid knowledge base of credit fraud with our current clients. We also have a unique history, having once been part of a large company before setting up on our own, which gives us a range of experience.
Question logo
What skills, experience, and expertise do you have onboard? What skill gaps do you have, and how are you addressing them?
We have the experience and knowledge base of a large company while also having the agility and innovation of a Fintech company. And perhaps most unusual for a Fintech, our first client was BNP Paribas, the largest bank in Europe. That speaks volumes because we design for that size and BNP's demands. Many startups at our level could not sustain these demands and are not as scalable. BNP has played an essential role in our history and helped us develop our technology to deliver our services today. We know that we need to hire more salespeople and strengthen our marketing team to increase our presence in the Fintech ecosystem.
Question logo
What are your biggest successes in the last six months?
Bleckwen has experienced an exciting hyper-growth period in the last few months. During this time, we have signed RCI (Renault Bank) in France and launched a fraud assessment with RCI in Spain following our proven success in France. We achieved our 2021 growth targets, doubled the size of the sales team, and expanded our marketing activities during this period.
Product
Question logo
How will you measure the accuracy of the model?
How will you measure the accuracy of the model?

We measure detectability (the fraud detected compared to all existing copies) and precision or relevance (the hustle saw compared to all alerts). These are the two key measures of accuracy. We can also take technical measurements if needed. Accuracy depends heavily on the constraints and thresholds set by the client.
Question logo
What threshold will you set for the model to determine if the model's performance is as expected?
That depends on the client because we build custom models based on your business constraints and tolerance for false positives. So the term 'as expected' depends on your needs and requirements. Our goal is to detect at least 50% of the remaining fraud.
Question logo
Do we sell to low-volume/high-value customers?
No. The models need enough data (24 to 36 months of historical data) to be well trained. If you are still doing manual checks to monitor your fraud, then our solution is not suitable for your needs at the moment. But in the meantime, we recommend you monitor on your historical data (e.g., suspected fraud, the definition of fraud in your business, distinguishing credit risk/loss to avoid mislabeling). This way, you can quickly implement our service once you have reached a sufficient volume.
Question logo
What is your ROI?
We target an ROI of 400%.
Question logo
Why should I not build this with my data science team?
There are multiple reasons why you should entrust the finding of fraud to Bleckwen rather than your internal data science team.
  1. Our data scientists are experts in building features and models to detect the weak signals inherent in fraud and have proven this to our other customers.
  2. Difficulty building, validating, and deploying the model to live can take time and money. We have perfected these processes by working with large banks, such as BNP.
  3. The costs of a data science team and difficulties in creating models can quickly add up.
Question logo
We use open banking functionality, so why do I need more checks?
AI is a powerful mathematical model for using business knowledge to make better decisions. If you are still not convinced, we are happy to offer our residual fraud assessment, which allows measuring your internal residual fraud along with a projection of the residual copy, we will reduce thanks to our credit service. This residual fraud assessment has a value of 30,000 euros, but we would like to offer it free. We are sure of the strength of our solution. So this assessment does not commit you to anything, to any resources.
Question logo
Can you advise our data science team?
As we are a services company, we do not see any opportunity to advise our data science team, but we would be happy to make some recommendations.
Question logo
Can the model be used for companies and individuals?
Our service can detect residual fraud on individuals, and we are currently working on detecting fraud in businesses.
Question logo
How does the service find forged documents?
We do not analyze the documents themselves, but we can find inconsistencies in the information provided in the raw data. In the application, some records are fake; others are genuine. So we can detect the fraud with fake documents without the papers themselves.
Question logo
What information does the dashboard display?
There are two types of dashboards in the credit solution. The first is the operational dashboard that the customer currently has access to, including alerts and usage statistics. The second is the ML monitoring dashboard, which we now only use.
Clients & Competitors
Question logo
How are you different from Onfido/Trans Union/Experian?
These companies use a generic model for fraud detection. Our solution is customized and complements their detection processes to precisely capture the fraud that passes through their system (residual deception).
Question logo
How does the credit solution differ from the competition?
Customized models with explainability. Proven ROI and fast integration.
Data
Question logo
In which country is our data processed (exactly)?
We process your data in France. If the interested party has other needs, we can adapt to his wishes to transfer them to another country.
Question logo
Will it be ok to send you all data in the anonymous form? If not, what data do you need to receive in their original format?
Yes, we accept all data in an anonymized format to comply with GDPR.
Question logo
What exactly do you want to be anonymized?
Specific anonymous data can affect the performance of the model.
Question logo
Are you GDPR compliant?
Yes, fully GDPR compliant - anonymous data.
Question logo
Do you do a document check?
We supplement existing detection systems (e.g. KYC checks where a document contains performance).
Question logo
How do we determine what fraud is?
That depends entirely on how the risk manager at your company defines fraud. Therefore, during our workshop with the risk manager, we must take some time to determine fraud in their company.
Question logo
Are you a processor or a controller? Do you use sub-processors?
We are a data processor. We do not use any sub-processors.
API
Question logo
Is this API able to monitor your customers for specific fraud typologies or fraud cases that occur in practice?
Right now, we focus on the onboarding phase - the loan application. However, we want to cover the entire loan lifecycle in the future. It would mean analyzing loan repayment data first to detect the signals of credit or fraud risk. It is a complex challenge for fraudsters looking to repay loans to access more credit and pause repayment across all loans. However, a challenge that we believe technology is best to solve.
Question logo
How do you implement your models in a decision engine environment?
We base our models on an XG boost system. It's a call-out and a real-time API.
Question logo
We want complete visibility and control over the models, so we have argued against APIs - what do you think?
The API allows for great flexibility in transferring your credit files and our scoring. In addition, with this API, you are always in control and have a complete overview. The API also saves you valuable time and resources because otherwise, you would have to develop your internal system to communicate specifically with us about credit applications. Again, this would cost you several months and resources (money invested and time spent by your technical team) to implement a solution that does not provide any additional benefit other than being built by your technical team. The advantage is that we only work with structured data that is anonymized.
Machine Learning Models
Question logo
Which models do you use?
We create custom models for our clients using their historical data to train.

It allows us to create a customized model for our client (e.g., the number of alerts generated, the number of fraud cases detected) because our goal is our client's satisfaction.
Question logo
How often do you retrain your models?
We monitor the models for variances, so it depends on the conflict - probably about every two months. Why you need to retrain - the drift issues arise when the model can no longer find new fraud patterns, as it is not incorporating these new patterns of fraud into its system. When the drift becomes apparent, we retrain the model. The frequency depends on the rise, so we monitor the importance, and if there is a noticeable drift, then the model needs to be retrained.
Question logo
How does your service compare against the cost to process and run internal detection systems?
Even if you already have an internal tool, 88% of AI projects do not produce. We will also provide you with a residual fraud assessment (24 to 36 months) based on your historical data. You will also get a projection of the residual fraud reduction using our offering.
Pricing
Question logo
What is your pricing model?
We operate on an annual subscription basis, and our business model is straightforward as we keep a percentage of what we save our customers in fraud. We base our pricing on the results of the residual fraud audit (our POC).